extra ops between loops?

Forum 2005-2010 (read only) Software Syntax & Programs
1

I wrote a simple code that does nothing but switch a pin on and off via port manipulations, with a one microsecond delay. I then hooked it up to an oscillator and noticed that between loops there seemed to be extra time being inserted.

#define VIDEO_ON PORTB|=B00000010
#define VIDEO_OFF PORTB&=B11111101

#define NOP asm("nop")

inline void one_us()
{
  // 1us / 62.5ns = 16
  NOP; NOP; NOP; NOP;
  NOP; NOP; NOP; NOP;
  NOP; NOP; NOP; NOP;
  NOP; NOP;
  //NOP; remove one to offset time for next instruction to complete
  //NOP; remove one more due to switching time, but will
  //result in slightly less than 1 us (but less than 62.5ns off)
  //~.996ns vs 1.06us, slightly less drift
}

void loop()
{

  VIDEO_ON;
  one_us();
  VIDEO_OFF;
  NOP;
}

If my information is correct, 1 nop should take 62.5ns, however in between the pin going low and high there's more than 62.5ns. The time between high and low is about 1us, as planned.

I expect that there's going to be a jump back to the top, which takes a couple of instructions, but even then I'm measuring ~800ns between loop end and start.

What's going on between loops? Is there anyway to cut down this extra time?

2

Well, in between the VIDEO_OFF and VIDEO_ON there is a call to a non-inline function, namely loop()!

What happens if you change loop() to:

void loop()
{
  while (true)
  {
    VIDEO_ON;
    one_us();
    VIDEO_OFF;
    NOP;
  }
}

?

Mikal

3

yes i realize that it calls loop() which takes some time, but it can't possibly take 800ns! that's at least 10 cycles.

interesting, I tried putting that while and it's dropped to 300ns. that's about 4~5 cycles, pretty reasonable for doing a while loop...

edit: I also tried modifying it to:

void loop()
{
  asm("VLoop:");
  VIDEO_ON;
  one_us();
  VIDEO_OFF;
  NOP;
  asm("jmp VLoop");
}

and got the same result (~300ns between pin going low and back high)

4

Hmm. I see:

000000a8 <loop>:
  a8:      29 9a             sbi      0x05, 1      ; 5
      ...
  c6:      29 98             cbi      0x05, 1      ; 5
  c8:      00 00             nop
  ca:      08 95             ret

000000cc <main>:
  cc:      0e 94 0e 01       call      0x21c      ; 0x21c <init>
  d0:      0e 94 54 00       call      0xa8      ; 0xa8 <loop>
  d4:      fd cf             rjmp      .-6            ; 0xd0 <main+0x4>

Note that "call" and "ret" are both 4-clock instructions, while rjmp is 2 clocks, so you "at least 10 cycles" is awfully close. sbi and cbi are also among the 2-cycle instructions.

If you're going to do video, you probably are going to need a more extensive set of tools than those provided by the arduino environment. The above dissassembly was obtained with "avr-objdump -d", which is included in the tools packaged with arduino (.../hardware/tools/avr/bin/avr-objdump on my mac. YMMV.)

(you also haven't done anything about the periodic timer tick, which will come along and take much more time, every once in a while (approximately every millisecond.))

5

FYI (I had it handy):

// disable timer 0 overflow interrupt
TIMSK0&=!(1<<TOIE0);
6

You must use bitwise negation (as opposed to logical negation).

TIMSK0&=~(1<<TOIE0);
7

thanks, forgot to consider that certain instructions take more cycles.

p.s. I've been doing cli(), does that actually disable interrupts?

8

I've been doing cli(), does that actually disable interrupts?

Yes but note that that method disables all interrupts. This may be an issue if you have serial I/O occurring at the same time and interrupts are disabled for more than 10 to 15 bit times.